| 报告简介 | In this work, we propose a new neural network framework to interface problems. The highlight of this framework lies in its incorporation of a neural network structure composed of multi-tensor neural networks and a loss function integrated with the Nitsche method. Meanwhile, we extend the application scope of tensor neural networks from computational regions that are hypercubes to that are unions of a finite number of disjoint hypercubes. We also propose a method for a special eigenvalue interface problem in engineering, the two-group neutron diffusion problem. We incorporates the idea of decoupling energy loss functions and changes the eigenvalue equations into fixed-source equations. As the neural network is optimized, the fixed-source term evolves with the iterative steps. Finally, ample numerical experiments are presented to validate our methods. |
